optimization for industrial problems978-3-642-24974-7/1.pdf · viii preface area of validity (c.f....
TRANSCRIPT
Optimization for Industrial Problems
123
Patrick Bangert
Optimization for IndustrialProblems
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)
liable to prosecution under the German Copyright Law.
and regulations and therefore free for general use.
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
or parts thereof is permitted only under the provisions of the German Copyright Law of September 9,1965, in its current version, and permission for use must always be obtained from Springer. Violations are
The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply,
reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publicationconcerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,
even in the absence of a specific statement, that such names are exempt from the relevant protective laws
© Springer-Verlag Berlin Heidelberg 2012
Patrick Bangertalgorithmica technologieBremen, Germany
s GmbH
Springer Heidelberg Dordrecht London New York
ISBN 978-3-642-24973-0 e-ISBN 978-3-642-24974-7DOI 10.1007/978-3-642-24974-7
Library of Congress Control Number:
Mathematics Subject Classification (2010):
2011945031
90-08, 90B50
It can be done !
algorithmica technologies GmbHAdvanced International Research Institute on Industrial
Optimization gGmbHDepartment of Mathematics, University College London
Preface
Some Early Opinions on Technology
There is practically no chance communications space satellites will be used to provide bettertelephone, telegraph, television, or radio service inside the United States
T. Craven, FCC Commissioner, 1961
There is not the slightest indication that nuclear energy will ever be obtainable. It wouldmean that the atom would have to be shattered at will.
Albert Einstein, 1932
Heavier-than-air flying machines are impossible.
Lord Kelvin, 1895
We will never make a 32 bit operating system.
Bill Gates, 1983
Such startling announcements as these should be deprecated as being unworthy of scienceand mischievous to its true progress.
William Siemens, on Edison’s light bulb, 1880
The energy produced by the breaking down of the atom is a very poor kind of thing. Anyonewho expects a source of power from the transformation of these atoms is talking moonshine.
Ernest Rutherford, shortly after splitting the atom for the first time, 1917
Everything that can be invented has been invented.
Charles H. Duell, Commissioner of the US Patent Office, 1899
Content and Scope
Optimization is the determination of the values of the independent variables in afunction such that the dependent variable attains a maximum over a suitably defined
vii
viii Preface
area of validity (c.f. the boundary conditions). We consider the case in which theindependent variables are many but the dependent variable is limited to one; multi-criterion decision making will only be touched upon.
This book, for the first time, combines mathematical methods and a wide rangeof real-life case studies of industrial use of these methods. Both the methods andthe problems to which they are applied as examples and case studies are useful inreal situations that occur in profit making industrial businesses from fields such aschemistry, power generation, oil exploration and refining, manufacturing, retail andothers.
The case studies focus on real projects that actually happened and that resulted inpositive business for the industrial corporation. They are problems that other com-panies also have and thus have a degree of generality. The thrust is on take-homelessons that industry managers can use to improve their production via optimizationmethods.
Industrial production is characterized by very large investments in technical fa-cilities and regular returns over decades. Improving yield or similar characteristicsin a production facility is a major goal of the owners in order to leverage their in-vestment. The current approach to do this is mostly via engineering solutions thatare costly, time consuming and approximate.
Mathematics has entered the industrial stage in the 1980s with methods suchas linear programming to revolutionize the area of industrial optimization. Neuralnetworks, simulation and direct modeling joined and an arsenal of methods nowexists to help engineers improve plants; both existing and new. The dot-com rev-olution in the late 1990s slowed this trend of knowledge transfer and it is safe tosay that the industry is essentially stuck with these early methods. Mathematics hasevolved since then and accumulated much expertise in optimization that remainshardly used. Also, modern computing power has exploded with the affordable par-allel computer so that methods that were once doomed to the dusty shelf can nowactually be used.
These two effects combine to harbor a possible revolution in industrial uses formathematical methods. These uses center around the problem of optimization asalmost every industrial problem concerns maximizing some goal function (usuallyefficiency or yield). We want to help start this revolution by a coordinated presenta-tion of methods, uses and successful examples.
The methods are necessarily heuristic, i.e. non-exact, as industrial problems aretypically very large and complex indeed. Also, industrial problems are defined byimprecise, sometimes even faulty data that must be absorbed by a model. Theyare always non-linear and have many independent variables. So we must focus onheuristic methods that have these characteristics.
This book is practical
This book is intended to be used to solve real problems in a handbook manner. Itshould be used to look for potential yet untapped. It should be used to see possibil-ities where there were none before. The impossible should move towards the realm
Preface ix
of the possible. The use, therefore, will mainly be in the sphere of application bypersons employed in the industry.
The book may also be used as instructional material in courses on either op-timization methods or applied mathematics. It may also be used as instructionalmaterial in MBA courses for industrial managers.
Many readers will get their first introduction as to what mathematics can reallyand practically do for the industry instead of general commonplaces. Many willfind out what problems exist where they previously thought none existed. Manywill discover that presumed impossibilities have been solved elsewhere. In total, Ibelieve that you, the reader, will benefit by being empowered to solve real problems.
These solutions will save the corporations money, they will employ people, theywill reduce pollution into the environment. They will have impact. It will showpeople also that very theoretical sciences have real uses.
It should be emphasized that this book focuses on applications. Practical prob-lems must be understood at a reasonable level before a solution is possible. Alsoall applications have several non-technical aspects such as legal, compliance andmanagerial ramifications in addition to the obvious financial dimension. Every so-lution must be implemented by people and the interactions with them is the principalcause for failure in industrial applications. The right change management includingthe motivation of all concerned is an essential element that will also be addressed.Thus, this book presents cases as they can really function in real life.
Due to the wide scope of the book, it is impossible to present neither the meth-ods nor the cases in full detail. We present what is necessary for understanding. Toactually implement these methods, a more detailed study or prior knowledge is re-quired. Many take-home lessons are however spelt out. The major aim of the bookis to generate understanding and not technical facility.
This book is intended for practitioners
The intended readership has five groups:
1. Industrial managers - will learn what can be done with mathematical methods.They will find that a lot of their problems, many seemingly impossible, are al-ready solved. These methods can then be handed to technical persons for imple-mentation.
2. Industrial scientists - will use the book as a manual for their jobs. They will findmethods that can be applied practically and have solve similar problems before.
3. University students - will learn that their theoretical subjects do have practicalapplication in the context of diverse industries and will motivate them in theirstudies towards a practical end. As such it will also provide starting points fortheses.
4. University researchers - will learn to what applications the methods that theyresearch about have been put or respectively what methods have been used byothers to solve problems they are investigating. As this is a trans-disciplinary
x Preface
book, it should facilitate communication across the boundaries of the mathemat-ics, computer science and engineering departments.
5. Government funding bodies - will learn that fundamental research does actuallypay off in many particular cases.
A potential reader from these groups will be assumed to have completed a math-ematics background training up to and including calculus (European high-school orUS first year college level). All other mathematics will be covered as far as needed.The book contains no proofs or other technical material; it is practical.
A short summary
Before a problem can be solved, it and the tools must be understood. In fact,a correct, complete, detailed and clear description of the problem is (measured intotal human effort) often times nearly half of the final solution. Thus, we will placesubstantial room in this book on understanding both the problems and the tools thatare presented to solve them.
Indeed we place primary emphasis on understanding and only secondary em-phasis on use. For the most part, ready-made packages exist to actually performan analysis. For the remainder, experts exist that can carry it out. What cannot bedenied however, is that a good amount of understanding must permeate the relation-ship between the problem-owner and the problem-solver; a relationship that oftenencompasses dozens of people for years.
Here is a brief list of the contents of the chapters
1. What is optimization?2. What is an optimization problem?3. What are the management challenges in an optimization project?4. How can we deal with faulty and noisy empirical data?5. How do we gain an understanding of our dataset?6. How is a dataset converted into a mathematical model?7. How is the optimization problem actually solved?8. What are some challenges in implementing the optimal solution in industrial
practice (change management)?
Most of the book was written by me. Any deficiencies are the result of my ownlimited mind and I ask for your patience with these. Any benefits are, of course,obtained by standing on the shoulders of giants and making small changes. Manycase studies are co-authored by the management from the relevant industrial cor-porations. I heartily thank all co-authors for their participation! All the case studieswere also written by me and the same comments apply to them. I also thank the co-authors very much for the trust and willingness to conduct the projects in the firstplace and also to publish them here.
Chapter 8 was entirely written by Andreas Ruff of Elkem Silicon Materials.He has many years of experience in implementing optimization projects’ results
Preface xi
in chemical corporations and has written a great practical account of the potentialpitfalls and their solutions in change management.
Following this text, we provide first an alphabetical list of all co-authors andtheir affiliations and then a list of all case studies together with their main topics andeducational illustrations.
Bremen, 2011 Patrick Bangert
Markus Ahorner algorithmica technologies GmbHCOO Gustav-Heinemann-Strasse [email protected] 28215 Bremen, GermanySection 4.8, p. 53; Section 4.9, p. 58 www.algorithmica-technologies.com
Dr. Patrick Bangert algorithmica technologies GmbHCEO Gustav-Heinemann-Strasse [email protected] 28215 Bremen, GermanyAll sections www.algorithmica-technologies.com
Claus Borgbohmer Sasol Solvents Germany GmbHDirector Project Management Romerstrasse [email protected] 47443 Moers, GermanySection 4.8, p. 53 www.sasol.com
Pablo Cajaraville Reiner MicrotekDirector Engineering and Sales Poligono Industrial Itziar, Parcela [email protected] 20820 Itziar-Deba, SpainSection 6.6, p. 135 www.reinermicrotek.com
Roger Chevalier EDF SA, R&D DivisionSenior Research Engineer 6 quai Watier, [email protected] 78401 Chatou Cedex, FranceSection 6.10, p. 152 www.edf.com
Jorg-A. Czernitzky Vattenfall Europe Warme AGPower Plant Group Director Berlin Puschkinallee [email protected] 12435 Berlin, GermanySection 7.10, p. 197 www.vattenfall.de
Prof. Dr. Adele Diederich Jacobs University Bremen gGmbHProfessor of Psychology P.O. Box 750 [email protected] 28725 Bremen, GermanySection 7.6, p. 183 www.jacobs-university.de
xii Preface
Bjorn Dormann Klockner Desma Schuhmaschinen GmbHResearch Director Desmastr. 3/[email protected] 28832 Achim, GermanySection 6.6, p. 135 www.desma.de
Hans Dreischmeier Vestolit GmbH & Co. KGDirector SAP Industriestrasse [email protected] 45753 Marl, GermanySection 4.9, p. 58 www.vestolit.de
Bernd Herzog Hella Fahrzeugkomponenten GmbHQuality Control Manager Dortmunder Strasse [email protected] 28199 Bremen, GermanySection 4.11, p. 63 www.hella.de
Dr. Philipp Imgrund Fraunhofer Institute for ManufacturingDirector Biomaterial Technology and Advanced Materials IFAMDirector Power Technologies Wiener Strasse [email protected] 28359 Bremen, GermanySection 6.6, p. 135 www.ifam.fhg.de
Maik Kohler Klockner Desma Schuhmaschinen GmbHTechnical Expert Desmastr. 3/[email protected] 28832 Achim, GermanySection 6.6, p. 135 www.desma.de
Lutz Kramer Fraunhofer Institute for ManufacturingProject Manager and Advanced Materials IFAMMetal Injection Molding Wiener Strasse [email protected] 28359 Bremen, GermanySection 6.6, p. 135 www.ifam.fhg.de
Guisheng Li Oil Production Technology Research InstituteInstitute Director Plant No. 5 of Petrochina Dagang Oilfield [email protected] Tianjin 300280, ChinaSection 6.12, p. 157 www.petrochina.com.cn
Bailiang Liu PetroChina Dagang Oilfield CompanyVice Director Tianjin [email protected] ChinaSection 7.9, p. 194 www.petrochina.com.cn
Preface xiii
Oscar Lopez MIM TECH ALFA, S.L.Senior Research Engineer Avenida Otaola, [email protected] 20600 Eibar, SpainSection 6.6, p. 135 www.alfalan.es
Torsten Mager KNG Kraftwerks- und Netzgesellschaft mbHDirector Technical Services Am Kuhlturm [email protected] 18147 Rostock, GermanySection 5.6, p. 102 www.kraftwerk-rostock.de
Manfred Meise Hella Fahrzeugkomponenten GmbHCEO Dortmunder Strasse [email protected] 28199 Bremen, GermanySection 4.11, p. 63 www.hella.de
Kurt Muller Vestolit GmbH & Co. KGDirector Maintenance Industriestrasse [email protected] 45753 Marl, GermanySection 4.9, p. 58 www.vestolit.de
Kaline Pagnan Furlan Fraunhofer Institute for ManufacturingResearch Assistant and Advanced Materials IFAMMetal Injection Molding Wiener Strasse [email protected] 28359 Bremen, GermanySection 6.6, p. 135 www.ifam.fhg.de
Yingjun Qu Oil Production Technology Research InstituteInstitute Director Plant No. 6 of Petrochina Changqing Oilfield Companyjyj [email protected] 718600 Shanxi, ChinaSection 6.12, p. 157 www.petrochina.com.cn
Pedro Rodriguez MIM TECH ALFA, S.L.Director R&D Avenida Otaola, [email protected] 20600 Eibar, SpainSection 6.6, p. 135 www.alfalan.es
Andreas Ruff Elkem Silicon MaterialsTechnical Marketing Manager Hochstadenstrasse [email protected] 50674 KlnChapter 8, p. 201 www.elkem.no
Dr. Natalie Salk PolyMIM GmbHCEO Am Gefach
xiv Preface
[email protected] 55566 Bad SobernheimSection 6.6, p. 135 www.polymim.com
Prof. Chaodong Tan China University of PetroleumProfessor Beijing [email protected] ChinaSection 6.12, p. 157; Section 7.9, p. 194 www.upc.edu.cn
Jorg Volkert Fraunhofer Institute for ManufacturingProject Manager and Advanced Materials IFAMMetal Injection Molding Wiener Strasse [email protected] 28359 Bremen, GermanySection 6.6, p. 135 www.ifam.fhg.de
Xuefeng Yan Beijing Yadan Petroleum Technology Co., Ltd.Director of Production Technology No. 37 Changqian Road, Changpingyxf [email protected] Beijing 102200, ChinaSection 6.12, p. 157 www.yadantech.com
Jie Zhang Yadan Petroleum Technology Co LtdVice CEO No. 37 Changqian Road, [email protected] Beijing 102200, ChinaSection 7.9, p. 194 www.yadantech.com
Timo Zitt RWE Power AGDirector Dormagen Combined-Cycle Plant Chempark, Geb. [email protected] 41538 DormagenSection 7.11, p. 199 www.rwe.com
Self-Benchmarking in Maintenance of a Chemical PlantSection 4.8, p. 53Summary: In addition to the common practice of benchmarking, we suggest to com-pare the plant to itself in the past to make a self-benchmark.Lessons: The right pre-processing of raw data from the ERP system can alreadybear useful information without further mathematical analysis.
Financial Data Analysis for Contract PlanningSection 4.9, p. 58
The following is a list of all case studies provided in the book. For each study, weprovide its location in the text and its title. The summary indicates what the casedeals with and what the result was. The “lessons” are the mathematical optimizationconcepts that this case particularly illustrates.
Preface xv
Summary: Based on past financial data, we create a detailed projection into thefuture in several categories and so provide decision support for budgeting.Lessons: Discovering basic statistical features of data first, allows the transformationof ERP data into a mathematical framework capable of making reliable projections.
Early Warning System for Importance of Production AlarmsSection 4.11, p. 63Summary: Production alarms are analyzed in terms of their abnormality. Thus weonly react to those alarms that indicate qualitative change in operations.Lessons: Comparison of statistical distributions based on statistical testing allowsus to distinguish normal from abnormal events.
Optical Digit RecognitionSection 5.4, p. 92Summary: Images of hand-written digits are shown to the computer in an effort forit to learn the difference between them without us providing this information (unsu-pervised learning).Lessons: It is possible to cluster data into categories without providing any infor-mation at all apart from the raw data but it pays to pre-process this data and to becareful about the number of categories specified.
Turbine Diagnosis in a Power PlantSection 5.5, p. 96Summary: Operational data from many turbines are analyzed to determine whichturbine was behaving strangely and which was not.Lessons: Time-series can be statistically compared based on several distinctive fea-tures providing an automated check on qualitative behavior of the system.
Determining the Cause of a Known FaultSection 5.6, p. 102Summary: We search for the cause of a bent blade of a turbine and do not find it.Lessons: Sometimes the causal mechanism is beyond current data acquisition andthen cannot be analyzed out of it. It is important to recognize that analysis can onlyelucidate what is already there.
Customer SegmentationSection 5.10, p. 117Summary: Consumers are divided into categories based on their purchasing habits.Lessons: Based on purchasing histories, it is possible to group customers into be-havioral groups. It is also possible to extract cause-effect information about whichpurchases trigger other purchases.
Scrap Detection in Injection Molding ManufacturingSection 6.6, p. 135
xvi Preface
Summary: It is determined whether an injection molded part is scrap or not.Lessons: Several time-series need to be converted into a few distinctive features tothen be categorized by a neural network as scrap or not.
Prediction of Turbine FailureSection 6.7, p. 140Summary: A turbine blade tear is correctly predicted two days before it happened.Lessons: Time-series can be extrapolated into the future and thus failures predicted.The failure mechanism must be visible already in the data.
Failures of Wind Power PlantsSection 6.8, p. 143Summary: Failures of wind power plants are predicted several days before they hap-pen.Lessons: Even if the physical system is not stable because of changing wind condi-tions, the failure mechanism is sufficiently predictable.
Catalytic Reactors in Chemistry and PetrochemistrySection 6.9, p. 148Summary: The catalyst deactivation in fluid and solid catalytic reactors is projectedinto the future.Lessons: Non-mechanical degradation can be predicted as well and allows for pro-jection over one year in advance.
Predicting Vibration Crises in Nuclear Power PlantsSection 6.10, p. 152Summary: A temporary increase in turbine vibrations is predicted several days be-fore it happens.Lessons: Subtle events that are not discrete failures but rather quantitative changesin behavior can be predicted too.
Identifying and Predicting the Failure of ValvesSection 6.11, p. 155Summary: In a system of valves, we determine which valve is responsible for a non-constant final mixture and predict when this state will be reached.Lessons: Using data analysis in combination with plant know-how, we can identifythe root-cause even if the system is not fully instrumented.
Predicting the Dynamometer Card of a Rod PumpSection 6.12, p. 157Summary: The condition of a rod pump can be determined from a diagram knownas the dynamometer card. This 2D shape is projected into the future in order todiagnose and predict future failures.
Preface xvii
Lessons: It is possible not only to predict time-series but also changing geometricalshapes based on a combination of modeling and prediction.
Human Brains use Simulated Annealing to ThinkSection 7.6, p. 183Summary: Based on human trial, we determine that human problem solving usesthe simulated annealing paradigm.Lessons: Simulated annealing is a very general and successful method to solve op-timization problems that, when combined with the natural advantages of the com-puter, becomes very powerful and can find the optimal solution in nearly all cases.
Optimization of the Muller-Rochow Synthesis of SilanesSection 7.8, p. 189Summary: A complex chemical reaction whose kinetics is not fully understood byscience is modeled with the aim of increasing both selectivity and yield.Lessons: It is possible to construct empirical models without theoretical understand-ing and still compute the desired answers.
Increase of Oil Production Yield in Shallow-Water Offshore Oil WellsSection 7.9, p. 194Summary: Offshore oil pumps are modeled with the aim of both predicting theirfuture failures and increasing the oil production yield.Lessons: The pumps must be considered as a system in which the pumps influenceeach other. We solve a balancing problem between them using their individual mod-els.
Increase of coal burning efficiency in CHP power plantSection 7.10, p. 197Summary: The efficiency of a CHP coal power plant is increased by 1%.Lessons: While each component in a power plant is already optimized, mathematicalmodeling offers added value in optimizing the combination of these componentsinto a single system. The combination still allows a substantial efficiency increasebased on dynamic reaction to changing external conditions.
Reducing the Internal Power Demand of a Power PlantSection 7.11, p. 199Summary: A power plant uses up some its own power by operating pumps and fans.The internal power is reduced by computing when these should be turned off.Lessons: We extrapolate discrete actions (turning off and on of pumps and fans)from the continuous data from the plant in order to optimize a financial goal.
Contents
1 Overview of Heuristic Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 What is Optimization? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Searching vs. Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.1.2 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.1.3 Finding through a little Searching . . . . . . . . . . . . . . . . . . . . . . 31.1.4 Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.1.5 Certainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Exact vs. Heuristic Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.2.1 Exact Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.2.2 Heuristic Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61.2.3 Multi-Objective Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Practical Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.4 Example Theoretical Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 Statistical Analysis in Solution Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.1 Basic Vocabulary of Statistical Mechanics . . . . . . . . . . . . . . . . . . . . . . 142.2 Postulates of the Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.3 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202.4 Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232.5 Ergodicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3 Project Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293.1 Waterfall Model vs. Agile Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303.2 Design of Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.3 Prioritizing Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4 Pre-processing: Cleaning up Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374.1 Dirty Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374.2 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.2.1 Time-Series from Instrumentation . . . . . . . . . . . . . . . . . . . . . . 384.2.2 Data not Ordered in Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
xix
xx Contents
4.3 Outlier Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404.3.1 Unrealistic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414.3.2 Unlikely Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414.3.3 Irregular and Abnormal Data . . . . . . . . . . . . . . . . . . . . . . . . . . 414.3.4 Missing Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.4 Data reduction / Feature Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434.4.1 Similar Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434.4.2 Irrelevant Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434.4.3 Redundant Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444.4.4 Distinguishing Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.5 Smoothing and De-noising . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474.5.1 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474.5.2 Singular Spectrum Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.6 Representation and Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504.7 Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514.8 Case Study: Self-Benchmarking in Maintenance of a Chemical Plant 53
4.8.1 Benchmarking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534.8.2 Self-Benchmarking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544.8.3 Results and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.9 Case Study: Financial Data Analysis for Contract Planning . . . . . . . . 584.10 Case Study: Measuring Human Influence . . . . . . . . . . . . . . . . . . . . . . . 624.11 Case Study: Early Warning System for Importance of Production
Alarms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5 Data Mining: Knowledge from Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 675.1 Concepts of Statistics and Measurement . . . . . . . . . . . . . . . . . . . . . . . . 67
5.1.1 Population, Sample and Estimation . . . . . . . . . . . . . . . . . . . . . 675.1.2 Measurement Error and Uncertainty . . . . . . . . . . . . . . . . . . . . 685.1.3 Influence of the Observer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 705.1.4 Meaning of Probability and Statistics . . . . . . . . . . . . . . . . . . . . 71
5.2 Statistical Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 735.2.1 Testing Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 735.2.2 Specific Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2.2.1 Do two datasets have the same mean? . . . . . . . . . . . 755.2.2.2 Do two datasets have the same variance? . . . . . . . . 765.2.2.3 Are two datasets differently distributed? . . . . . . . . . 765.2.2.4 Are there outliers and, if so, where? . . . . . . . . . . . . . 775.2.2.5 How well does this model fit the data? . . . . . . . . . . . 78
5.3 Other Statistical Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 795.3.1 Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 795.3.2 ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 815.3.3 Correlation and Autocorrelation . . . . . . . . . . . . . . . . . . . . . . . . 845.3.4 Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 855.3.5 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 895.3.6 Fourier Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Contents xxi
5.4 Case Study: Optical Digit Recognition . . . . . . . . . . . . . . . . . . . . . . . . . 925.5 Case Study: Turbine Diagnosis in a Power Plant . . . . . . . . . . . . . . . . . 965.6 Case Study: Determining the Cause of a Known Fault . . . . . . . . . . . . 1025.7 Markov Chains and the Central Limit Theorem . . . . . . . . . . . . . . . . . . 1055.8 Bayesian Statistical Inference and the Noisy Channel . . . . . . . . . . . . 107
5.8.1 Introduction to Bayesian Inference . . . . . . . . . . . . . . . . . . . . . . 1075.8.2 Determining the Prior Distribution . . . . . . . . . . . . . . . . . . . . . . 1085.8.3 Determining the Sampling Distribution . . . . . . . . . . . . . . . . . . 1105.8.4 Noisy Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.8.4.1 Building a Noisy Channel . . . . . . . . . . . . . . . . . . . . . 1115.8.4.2 Controlling a Noisy Channel . . . . . . . . . . . . . . . . . . . 112
5.9 Non-Linear Multi-Dimensional Regression . . . . . . . . . . . . . . . . . . . . . 1135.9.1 Linear Least Squares Regression . . . . . . . . . . . . . . . . . . . . . . . 1135.9.2 Basis Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1145.9.3 Nonlinearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.10 Case Study: Customer Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6 Modeling: Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1216.1 What is Modeling? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.1.1 Data Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1246.1.2 How much data is enough? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.2 Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1266.3 Basic Concepts of Neural Network Modeling . . . . . . . . . . . . . . . . . . . 1296.4 Feed-Forward Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1316.5 Recurrent Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1326.6 Case Study: Scrap Detection in Injection Molding Manufacturing . . 1356.7 Case Study: Prediction of Turbine Failure . . . . . . . . . . . . . . . . . . . . . . 1406.8 Case Study: Failures of Wind Power Plants . . . . . . . . . . . . . . . . . . . . . 1436.9 Case Study: Catalytic Reactors in Chemistry and Petrochemistry . . . 1486.10 Case Study: Predicting Vibration Crises in Nuclear Power Plants . . . 1526.11 Case Study: Identifying and Predicting the Failure of Valves . . . . . . . 1556.12 Case Study: Predicting the Dynamometer Card of a Rod Pump . . . . 157
7 Optimization: Simulated Annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1657.1 Genetic Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1667.2 Elementary Simulated Annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1677.3 Theoretical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1697.4 Cooling Schedule and Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
7.4.1 Initial Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1737.4.2 Stopping Criterion (Definition of Freezing) . . . . . . . . . . . . . . 1747.4.3 Markov Chain Length (Definition of Equilibrium) . . . . . . . . . 1757.4.4 Decrement Formula for Temperature (Cooling Speed) . . . . . 1777.4.5 Selection Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1787.4.6 Parameter Choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
7.5 Perturbations for Continuous and Combinatorial Problems . . . . . . . . 181
xxii Contents
7.6 Case Study: Human Brains use Simulated Annealing to Think . . . . . 1837.7 Determining an Optimal Path from A to B . . . . . . . . . . . . . . . . . . . . . . 1867.8 Case Study: Optimization of the Muller-Rochow Synthesis of
Silanes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1897.9 Case Study: Increase of Oil Production Yield in Shallow-Water
Offshore Oil Wells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1947.10 Case Study: Increase of coal burning efficiency in CHP power plant 1977.11 Case Study: Reducing the Internal Power Demand of a Power Plant 199
8 The human aspect in sustainable change and innovation . . . . . . . . . . . . 2018.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
8.1.1 Defining the items: idea, innovation, and change . . . . . . . . . . 2028.1.2 Resistance to change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
8.2 Interface Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2078.2.1 The Deliberate Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . 2078.2.2 The Healthy Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
8.3 Innovation Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2138.4 Handling the Human Aspect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
8.4.1 Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2178.4.2 KPIs for team engagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2198.4.3 Project Preparation and Set Up . . . . . . . . . . . . . . . . . . . . . . . . . 2218.4.4 Risk Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2238.4.5 Roles and responsibilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2268.4.6 Career development and sustainable change . . . . . . . . . . . . . . 2288.4.7 Sustainability in Training and Learning . . . . . . . . . . . . . . . . . . 2318.4.8 The Economic Factor in Sustainable Innovation . . . . . . . . . . . 232
8.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243